Agentic Commerce: How AI Agents Transact On-Chain
Agentic commerce is AI agents discovering, negotiating and paying for goods and services autonomously. This guide compares the major 2030 forecasts, walks through the machine-to-machine payment flow step by step and explains why on-chain rails fit agent transactions.
7 min read
35 views
Key takeaways: agentic commerce and on-chain payments 4
What the major 2030 forecasts say, how a machine-to-machine payment flow actually works and why on-chain rails fit agent transactions.
- Forecasts diverge but agree on direction 2030 estimates range from Bain's $300-500B of US eCommerce to McKinsey's up to $1T US orchestrated retail revenue and $3-5T globally.
- Six-step machine-to-machine flow Mandate, discovery, agreement, payment, settlement and audit trail replace the human checkout sequence end to end.
- On-chain rails fit agent speed and scale Programmable authorization, machine-speed settlement and micropayment economics push agent-to-agent commerce toward stablecoin rails.
- Mandate design comes before payment integration Expressing what an agent may buy precisely enough to enforce is the hard problem, not wiring up the payment rail.
Retail and B2B commerce have run on the same assumption for decades: a person makes the buying decision and clicks pay. That assumption is now breaking down as businesses delegate discovery, comparison and purchase to AI agent development projects built to act on their behalf. This guide compares the major 2030 forecasts for agentic commerce, walks through the machine-to-machine payment flow step by step and explains why on-chain rails fit agent transactions better than the checkout page they are replacing.
In short: agentic commerce is when AI agents complete purchases on a user’s behalf: they discover products, compare options, negotiate and pay without a human clicking checkout. Analyst forecasts for 2030 range from Bain’s $300-500B of US eCommerce to McKinsey’s up to $1T of US orchestrated retail revenue and $3-5T globally. Because agents transact machine-to-machine at machine speed, programmable on-chain rails, stablecoins with spending mandates encoded in smart contracts, are emerging as a natural settlement layer for them.
What agentic commerce actually is
Traditional eCommerce assumes a human at every decision point: search, compare, add to cart, enter card details. Agentic commerce removes the human from the loop for delegated tasks. You tell an agent “keep the office stocked with coffee under $200 a month” or “book the cheapest compliant freight for this shipment” and the agent handles discovery, selection and payment within the boundaries you set.
This stopped being theoretical in late 2025. Adobe measured an 805% year-over-year increase in AI-driven traffic to US retail sites on Black Friday 2025, attributed roughly $3B of US Black Friday sales to AI agents and estimated $14.2B in global GenAI and agent-driven online sales for 2025 (Adobe, 2025). Those are small numbers against total eCommerce, but the growth rate is what has every payments and retail strategy team paying attention.
The infrastructure is arriving in parallel: agent-payment protocols from major payment and platform players, merchant-side agent interfaces and, the focus of this article, on-chain settlement rails designed for autonomous, machine-to-machine transactions.
What the forecasts say about 2030
The major houses disagree on magnitude but not on direction. The table below compares the published 2030 projections. We deliberately exclude figures we could not verify against a primary source.
| Source | 2030 projection | Scope |
|---|---|---|
| McKinsey | Up to $1T US orchestrated retail revenue; $3-5T globally | Retail revenue orchestrated by AI agents |
| Bain | $300-500B in the US (15-25% of eCommerce) | US eCommerce transacted via agents |
| Morgan Stanley | 10-20% of US eCommerce agent-driven ($190-385B) | US eCommerce share |
| J.P. Morgan | Up to 25% of US online sales | US online sales share |
Read together: the consensus corridor is that between one tenth and one quarter of US online commerce runs through agents by 2030, with McKinsey’s global orchestration estimate an order of magnitude larger because it counts influence over the purchase, not just execution.
On the payments-infrastructure side specifically, Galileo projects the agentic payments market growing 13x to $93B by 2032 (Galileo, 2026). Whichever number proves right, payment flows designed for human checkout pages will not carry this volume unchanged.
How a machine-to-machine payment flow works
An agent buying from another agent (or from a merchant’s automated storefront) follows a flow that looks different from card checkout at every step. Here is the canonical sequence.
Step 1: mandate and delegation
The human principal grants the agent a spending mandate: scope (what it may buy), budget (how much), counterparty rules (from whom) and time bounds. On-chain, this becomes a programmable authorization, a smart contract or session key that cryptographically enforces the limits rather than trusting the agent’s code to respect them. This is where smart contract development work concentrates in most agentic commerce engagements.
Step 2: discovery and negotiation
The buying agent queries catalogs, APIs or other agents for offers. Because both sides are software, negotiation can be genuinely dynamic: quantity discounts, delivery windows and service levels resolved in milliseconds through structured offer-counteroffer exchanges rather than static price pages.
Step 3: agreement and intent
The agents converge on terms and produce a signed order: a machine-readable commitment specifying goods, price, settlement asset and delivery conditions. Signing matters, each side needs non-repudiable proof of what was agreed, because no human saw the screen.
Step 4: payment execution
The buying agent triggers payment within its mandate. On-chain this is typically a stablecoin transfer, either direct or into an escrow contract that releases funds when delivery conditions are met. The mandate contract verifies the transaction against its limits before signing: right counterparty, within budget, inside the time window.
Step 5: settlement and receipt
The transfer finalizes on-chain in seconds and both agents receive a cryptographic receipt, the transaction hash plus the signed order, which flows into each principal’s ledger for reconciliation, accounting and dispute evidence.
Step 6: audit trail
Every step above is logged. For businesses this is the non-negotiable part: when an agent spends company money, finance and compliance need a complete, tamper-evident record of the mandate, the decision inputs and the settlement.
Why on-chain rails fit agent payments
Nothing stops agents from using cards, and card networks are building agent tokens for exactly that. But several properties push machine-to-machine commerce toward crypto payments rails:
- Programmable authorization. Spending limits enforced by a smart contract are stronger than limits enforced by the agent’s own code. The mandate is the rail, not a policy layered on top of it.
- Machine-speed settlement. Agents transact 24/7 and may chain purchases (buy compute, then data, then bandwidth) within seconds. Final settlement in seconds beats T+1 batching.
- Micropayments. Agent-to-agent economies involve tiny transactions, paying per API call, per inference, per data record. Low-fee networks make sub-cent payments economically viable where card economics never could.
- Native machine identity. A wallet address plus verifiable credentials gives an agent an economic identity without a card issuer deciding whether software can be a cardholder.
- Escrow without intermediaries. Conditional payment (release on delivery) is a native smart contract pattern, replacing the dispute machinery that card rails need because payment and fulfillment are otherwise disconnected.
The realistic 2026 picture is hybrid: card networks handle agent purchases from traditional merchants while on-chain stablecoin rails handle agent-to-agent and high-frequency machine commerce. Enterprises building agents should design for both.
What builders should get right
Three engineering decisions dominate agentic commerce projects in our experience building AI agents and payment infrastructure:
- Mandate design before payment integration. The hard problem is expressing “what this agent may buy” precisely enough to enforce. Start from the authorization model and let the payment solutions development work follow.
- Human checkpoints by value, not by transaction. Approving every purchase kills the point of delegation. Approving nothing invites disaster. Set escalation thresholds by amount, novelty of counterparty and deviation from past behavior.
- Reconciliation as a first-class system. Autonomous spending only survives contact with a CFO if every on-chain settlement maps cleanly to an order, a budget line and an accounting entry.
Security deserves its own article, agents that spend money are agents worth attacking and the fraud patterns emerging around agent workflows are covered in our companion piece on fraud in the agentic era.
How Pharos Production builds agentic commerce infrastructure
Agentic commerce moves the buyer from a person to a program, and that changes what a payment rail must provide: enforceable mandates, machine-speed settlement, micropayment economics and audit trails no human ever has to screenshot. On-chain stablecoin rails supply those properties natively, which is why the forecast curves for agentic commerce and programmable payments are bending upward together.
If you are building agents that need to transact, or payment infrastructure that needs to serve them, Pharos Production designs and ships both. Start with our AI agent development team to scope an agent architecture with payments, mandates and auditability built in from day one.
Sources: 2030 forecast figures from McKinsey, Bain, Morgan Stanley and J.P. Morgan as published in their respective 2025-2026 agentic commerce research, Adobe’s 2025 Black Friday and GenAI shopping data and Galileo’s 2026 agentic payments market projection. Figures are analyst estimates, not guarantees, and the houses disagree on scope and magnitude.
FAQ
Quick answers to common questions about custom software development, pricing, process and technology.
Type to filter questions and answers. Use Topic to narrow the list.
Showing all 3
No matches
Try a different keyword, change the topic, or clear filters
-
Agentic commerce is commerce where AI agents execute purchases autonomously on behalf of a person or business: discovering products, comparing offers, negotiating terms and completing payment within a mandate the principal defines. Analysts project it will carry a meaningful share of online commerce by 2030, Morgan Stanley estimates 10-20% of US eCommerce, J.P.
Morgan up to 25% of US online sales.
-
Through two families of rails. Card networks are extending tokenized card credentials to agents for purchases from traditional merchants.
On-chain rails let agents hold stablecoins in programmable wallets where a smart contract enforces the spending mandate: counterparty, budget and time limits are checked cryptographically before any transfer executes, and settlement finalizes in seconds with a verifiable receipt.
-
Published forecasts diverge but agree on direction: Bain projects $300-500B of US eCommerce (15-25%), Morgan Stanley 10-20% of US eCommerce ($190-385B), J.P. Morgan up to 25% of US online sales and McKinsey up to $1T of US orchestrated retail revenue with $3-5T globally.
Early signals support the trajectory: Adobe attributed roughly $3B of US Black Friday 2025 sales to AI agents after an 805% year-over-year jump in AI-driven retail traffic.
I work with startup founders who need a dedicated software development team but don’t want to gamble on hiring, random outsourcing, or opaque delivery.
Most founders face the same problem sooner or later.
Early technical and team decisions lock the product into tech debt, slow delivery, missed milestones and constant re-hiring. By the time this becomes visible, fixing it is already expensive.As a CTO and software architect, I help founders design, build and run dedicated development teams that work as a true extension of the startup. Not as a black-box vendor.
My focus is on complex products where mistakes are costly:
- Web3 and blockchain platforms
- FinTech and regulated products
- High-load startup systems
- MVP → scale transitions
We don’t do body-shopping.
We don’t sell generic outsourcing.Instead, we help founders:
- build the right team structure from day one
- keep technical ownership and transparency
- scale delivery without losing control
- avoid vendor lock-in and hidden risks
Teams are aligned with the product roadmap, business goals and long-term architecture. Not just short-term velocity.
